Description
If the purpose of a slavedelayed secondary is to facilitate "going back" to a previous database state, then slavedelayed secondaries are of little value in a sharded cluster in general (i.e., without placing constraints on the operation of the cluster, e.g., that the balancer hasn't run lately).
The docs ought to note this fact.
Whether the docs describe various edge cases where they might be useful is outside the scope of this request, though I assign little value to documenting marginal edge cases or convoluted operational practices.